15th ICPhS Barcelona
KACST Arabic Phonetics Database
Mansour M. Alghmadi
Computer & Electronics Research Institute,
King Abdulaziz City for Science and Technology,
Riyadh, Kingdom of Saudi Arabia
mghamdi@hotmail.com
ABSTRACT
Arabic sounds were the first to be fully described and
analyzed. The place and manner of articulation of each
Arabic sound were identified and documented in the eighth
century AD by Sibawayah in his famous treatise al-Kitaab.
Since then many Arabic scholars have quoted him, but not
much has been added to what he wrote.
Lately, King Abdulaziz City for Science and Technology
(KACST) has published a detailed and comprehensive
database called KACST Arabic Phonetics Database
(KAPD) [1]. KAPD gives almost all the details of the
articulatory mechanism of Arabic sounds. It contains more
than 46000 files. They show the results of 9 experiments on
8 native Arabic subjects. The experiments include: airflow,
air pressure, linguapalatal contact, nasality, perception, side
and front facial images, and stroboscopic images of the
glottis, pharyngeal cavity and velo-pharyngeal port. The
data in KAPD are raw, i.e., it can be utilized for research
and development such as speech therapy, speech perception,
speech synthesis, speech recognition and speech modeling.
KAPD is available on 3 CD's for researchers and students
of Speech.
1. INTRODUCTION
Arabic sounds were identified and their places and manners
of articulation were described and categorized 13 centuries
ago by Sibawayh in his treatise Al-Kitab [2]. Many other
Arab linguists came after him and wrote on Arabic sounds
but Al-Kitab has remained their main reference. However,
since the birth of Phonetics as an independent area of
science in the 19th century and the use of modern equipment
and research tools, the studies of Arabic sounds remain
mostly individual efforts by Ph. D. students and a few
interested phoneticians on certain aspects of Arabic sounds.
Their results on articulatory phonetics are reported in
references such as [3, 4, 5, 6, 7, 8, 9, 10 and 11]. Other
results on Arabic acoustic phonetics are available in [6, 11,
12, 13, 14, 15 and 16], to mention some. A few researchers
have worked on Arabic speech recognition [9 and 15].
Researchers usually collect data, analyze them, study them
and write their findings on papers, as it is the case in almost
all research fields. What other researchers would like to do
is take the results and try to utilize them in other related
3109
areas of knowledge, but they do not have access to the
original data. The original data are some times of an
important value for certain researchers who would like to
study them from different point of view or would like to
utilize them in areas such as training a speech recognition
engine or modeling a text-to-speech system.
King Abdulaziz City for Science and Technology (KACST)
has realized the need for a database on Arabic phonetics
which can be used by researchers of Arabic in particular
and human speech in general. In 2000, KACST introduced
its Arabic Phonetics Database (KAPD). The database
includes more than 46000 files on Arabic sounds. The files
are photos taken by a laryngoscope and video cameras
(JPEG format), NSP file formats acquired by some of Kay's
software and hardware systems, audio files in .wav format,
and a table of the perceptual responses to all uttered tokens.
2. METHOD
The objective of KAPD was to provide researchers who are
interested in Arabic sounds with as much detailed data as
possible. Therefore, 9 experiments were designed to collect
data for each sound: 1) Facial images, 2) velo-pharyngeal
port images, 3) pharyngeal images, 4) laryngeal images, 5)
aerodynamic data, 6) linguapalatal-contact data, 7)
electroglottographic (EGG) data, 8) nasality data 9)
perceptual results.
The tokens in experiments 2, 3, 4 and 5 are in the frame
cvCvc/cvCCvc where c is /z/, v is /a/ and C is one of the
Arabic 28 consonants. The target consonant is single in
cvCvc and geminate in cvCCvc. The tokens in experiments
1, 6, 7 and 8 are in the frames: cvCvc/cvCCvc, Cvc and cvC
where c is /z/, v is one of the vowels /a , i , u/ and C is one of
the Arabic 28 consonants. The target consonant C is single
in cvCvc and geminate in cvCCvc. The Arabic consonants
which were used in the experiments are: the plain stops /b ,
t , d , k , q , / /; the emphatic stops /t÷ , d÷ /; the plain
fricatives /P , D , f , s , z , S , Z , X , “ ,  , ÷ , h/; the emphatic
fricatives /D÷ , s÷ /; the nasals /m , n/; the glides /j , w/; the
lateral /l/; and the trill /r/.
7 subjects participated in the experiments 1 to 8 (Table 1),
while 10 other subjects participated in the perception
experiment 9. All the subjects are native speakers of Arabic
and carry the Saudi accent of Arabic. The subjects in the
perception experiment are undergraduate students at
ISBN 1-876346-48-5 © 2003 UAB
15th ICPhS Barcelona
Umalqura University and none of them is one of the 7
subjects in the other experiment.
Average SD
Age 32.71 years 3.73
Height 167.71 cm 4.11
Weight 77.57 kg 11.66
Table 1. The average and standard deviation
(SD) of the subjects' age height and weight.
In experiment 1, a color video camera (Sony Model:
DXC-637P) was used for side view and another one
(Sony Model: BVP-7P) was used for the front view.
The recording was made by a video cassette recorder
(Model: BVW-75P BETACAM SP). In experiment 2,
3 and 4, a Rhino-Laryngeal stroboscope (Model: RLS
9100) was used (Figure 1). In experiment 5, an
aerophone (Model: Aerophone-II 6800) was used to
collect and record the data (Figure 2). In experiment 6,
Palatometer (Model: 6300) was used (Figure 3). In
experiment 7, ElectroGlottoGraph was used. In
experiment 8, Nasometer (Model: 6200-3) was used. In
all the above experiments CSL (Computerized Speech
Laboratory, Model: 4300B) was used to record the
acoustic signal of the subjects' utterances. Most of the
equipment used in these experiments carries the name of
Kaytelemetrics. However, some of them were modified
to cover the need of the experiments and the Arabic
sounds (for example, Figure 3). In experiment 9, a
computer program was written to give the subjects the
choice of selecting the target sound that they have heard,
its position in the token and its adjacent vowel.
Figure 1. The three circles show the position of the distal
tip in experiments 2, 3 and 4.
Figure 2. The distal tip of the laryngoscope, the
air-presser tube (the measurement is in centimeters), and
one of the subjects wearing the air flew mask while the
air-presser tube is inserted into the pharyngeal cavity via
the nasal cavity.
ISBN 1-876346-48-5 © 2003 UAB
Figure 3. The front and side views of the modified artificial
palate with an extension to cover the uvula area.
3. RESULTS
The results of experiment 1 are under directory C in KAPD.
They include facial images of the 7 subjects during the
production of Arabic sounds (Figure 4), NSP files with tags
showing the target sound onset and offset, and WAV files of
the same tokens. The NSP files contain the acoustic data
plus other data such as linguapalatal contact and EGG. NSP
files can be opened by a CSL or Multi-Speech program,
both produced by Kaytelemetrics. WAV files can be opened
by many programs including Windows and CSL. The
results of experiment 2 are under the directory V. They
include images of the velopharyngeal port (Figure 5), NSP
and WAV files. The results of experiment 3 are under the
directory E. The results of this experiment include images
of the pharyngeal cavity (Figure 5), NSP and WAV files.
The results of experiment 4 are under the directory X. They
include images of the glottis (Figure 5), NSP and WAV files.
The results of experiment 5 are under Directory A. They
include NSP and WAV files. The NSP files have the
aerodynamic data (Figure 6). The results of Experiment 6
are under Directory P. They include NSP (Figure 7) and
WAV files. The results of experiment 7 are under Directory
G. They include NSP (Figure 8) and WAV files. The results
of experiment 8 are under Directory N. They include NSP
(Figure 9) and WAV files. The results of experiment 9 are in
the File DB1. They are the responses of the 10 informants
who listened to the tokens produced by the subjects in the
first 8 experiments and made their judgments on the target
sounds they had heard.
Figure 4. The side and front views of the face of one of
the subjects.
3110

Figure 5. Images of, from left to right, the vocal folds
(experiment 2), epiglottis (experiment 3) and the
velopharyngeal port (experiment 4).
Figure 6. Window 1 shows the waveform of /zakkaz/,
window 2 shows the airflow and window 3 shows the air
pressure of the same token (experiment 5).
Figure 7. Window 1 shows the waveform of /zaXXaz/,
window 2 shows the linguapalatal contact during the
production of the target sound /X/ (experiment 6).
Figure 8. Window 1 shows the waveform of /zakkaz/,
window 2 shows the laryngeal activities and window 3
magnifies a portion of window 2 (experiment 7).
3111
15th ICPhS Barcelona
Figure 9. Window 1 shows the waveform of /zammaz/,
window 2 shows the oral air pressure and window 3
shows the nasal air pressure (experiment 8).
The name of each file in the database has 7 letters. The first
letter from left represents the experiment, the second
represents the subject, the third and forth represent the
target sound, the fifth represents the gemination, the sixth
represents the position of the target sound in the token, and
the seventh represents the adjacent vowel to the target
sound. For example, the file PMBSSIU.NSP means that
this file is about experiment 6 (linguapalatal contact), M is
the subject's initial, BS represent the /b/ sound, S means
the target sound is single not geminate, I means the target
sound is in the initial position of the token, U means the
adjacent vowel is /u/. The extension NSP is the file format.
4. CONCLUSIONS
This paper introduces the reader to a database on Arabic
sounds produced by Saudi subjects. The database includes
articulatory, acoustic and perceptual information. The
articulatory data consist of images of the face,
velopharyngeal port, pharyngeal cavity and glottis. They
also include linguapalatal contacts and aerodynamic data.
The acoustic data consist of the audio recording of the
tokens made by the subjects in all experiments. The
perceptual data show the responses of 10 informants to the
Arabic sounds in the acoustic data.
KAPD can be utilized in various areas. To mention some, in
speech-to-text experiments, the acoustic part of the
database can be used to train and test the system where the
recording of 7 native Arabic speakers producing Arabic
sounds 8 times in different token positions. KAPD can be
used in text-to-speech systems especially when using the
concatenation method. It can also be used in vocal tract
modeling and facial animation during speech. It would be
of value for researchers in phonetics, speech therapy and
forensic phonetics.
5. ACKNOWLEDGEMENT
This paper and the database project are sponsored by King
Abdulaziz City for Science and Technology. The database
team consists of: Mansour Alghamdi, Ashraf Alkhairi,
Abdullah Azzamil, Muhammad Alfarraj, Muhammad
Aljabr, Muhammad Basalamah, Sayed Hussain, Surraya
Addawsari, and Waqayyan Addawsari.
ISBN 1-876346-48-5 © 2003 UAB
15th ICPhS Barcelona

REFERENCES [15] Alghamdi, M. (1990) Analysis, synthesis and

perception of Arabic voicing; Ph. D. dissertation;
[1] (2000) KACST Arabic Phonetics Database, Computer University of Reading.
and Electronics Research Institute, King Abdulaziz
City for Science and Technology, Riyadh. [16] Alghamdi, M. (1998) A spectrographic analysis of
Arabic vowels: a cross-dialect study, J. King Saud Univ.
[2] Nassir, Abdulmun'im (1993) Sibawayh the 10, 3-24.
Phonologist: A Critical Study of the Phonetic and
Phonological Theory of Sibawayh As Presented in His
Treatise Al-Kitab, Kegan Paul Intl.
[3] Kuriyagawa, F., M. Sawashima, et al. (1988).
“Electromyographic study of emphatic consonants in
standard Jordanian Arabic”, Folia Phoniatrica:
International Journal of Phoniatrics, Speech Therapy
and Communication Pathology 40(3): 117-122.
[4] Giannini, A. (1982). The Emphatic Consonants in
Arabic. Naples, Istituto Univ. Orientale, Seminario di
Studi dell'Europa Orientale, Laboratorio di Fonetica
Sperimentale.
[5] Gegov, K. (1989). “The Bulgarian velar consonant /x/
and three Arabic fricative consonants.” Contrastive
Linguistics 14(2): 21-24
[6] Butcher, A. and K. Ahmad (1987). “Some acoustic and
aerodynamic characteristics of pharyngeal consonants
in Iraqi Arabic.” Phonetica 44(3): 156-172.
[7] Boff, M.-C. (1983). “A Contribution to the
experimental study of back consonants in classical
Arabic (Moroccan speakers).” Travaux de l'Institut de
Phonetique de Strasbourg 15: 1-363.
[8] Ghali, M. M. (1983). “Pharyngeal articulation
(Especially in Arabic and English).” Bulletin of the
School of Oriental and African Studies-University of
London 46(3): 432.
[9] Alwan, A. (1989). “Perceptual cues for place of
articulation for the voiced pharyngeal and uvular
consonants.” Journal of the Acoustical Society of
America 86(2): 549-556.
[10] Ghazali, S.; (1977); Back consonants and backing
coarticulation in Arabic; Ph. D. dissertation; University
of Texas at Austin.
[11] Al-Ani, S. (1970) Arabic Phonology. The Hague.
[12] El-Halees, Y. A. (1987). “F2 and back articulations in
Arabic.” Journal of the College of Arts, King Saud
University 14(1): 21-44.
[13] El-Halees, Y. (1985). “The role of F1 in the
place-of-articulation distinction in Arabic.” Journal of
Phonetics 13(3): 287-298.
[14] Flege, J., and Port, R. (1981) Cross-language phonetic
interference: Arabic to English, Language and Speech
24, 125-146.

ISBN 1-876346-48-5 © 2003 UAB 3112